Macbook integrated vs discrete gpu. No gain.

[0x80]

I am about to return my macbook to my previous employer, and so I need to decide on what model to replace it with. My current model is late 2013 with the 750M discrete graphics. I don't care about 3d performance much and so I was interested to see if there is much difference between the discrete Intel Iris Pro graphics and the 740M in terms of video performance.

I did a benchmark which I posted here:
viewtopic.php?f=11&t=11093&start=10&hil ... ark#p45349

From my test I like to conclude that there is no noticeable performance gain in Resolume by using discrete graphics

At first, looking at Oaktown's results, I noticed that CPU makes more of a difference, since he reported 13 layers of clean instead of my 9. But then I read gerbyzation's benchmark which has the exact same specs as my computer, but with an older version of Resolume, and he reports 10 layers. Also 10 layers of noise which is the same as in Oaktown's results.

Based on my findings I would get a macbook without the discrete graphics but with an upgraded cpu maybe. But if you need the 16gb of ram and 512gb of storage like me, you might as well, since then the 750M then comes for free basically...

I think it's interesting to know that for Resolume it doesn't matter if you're using the discrete or Iris Pro graphics. Having a model without discrete graphics could then be a benefit since you'll get lower power consumption as it won't switch to discrete for applications and driving external displays.

So once apple starts putting 16 ram or 512 storage in the lower range models I would prefer them. Too bad I can't hold out for the next models.

For additional information on the comparison between Iris Pro and 750M I found this article to be a good read:
http://www.archagon.net/a-few-pointless ... acbook-pro

I was aware that Windows has better graphics drivers, but the difference in these benchmarks really surprise me. I wonder how much of that can be blamed on drivers or optimisations in Resolume itself. I hope that OSX will improve some in this area with Yosemite.

I am interested to hear your thoughts on this and if you think I am mistaken somewhere.

[Oaktown]

Don't even think about it, if you can afford it, get the Macbook Pro with the GT 750M card!

As I mentioned in your other post (viewtopic.php?f=11&t=11093&p=45351#p45351) Resolume is all about being able to show your work on external displays and trust me you don't want to be running shows on an integrated GPU. The Intel Iris Pro Graphics dynamically allocates up to 1 GB of system memory whereas the GT 750M has 2GB of dedicated GDDR5 memory.

Resolume used with the DXV codec (which as a side note is what you should always use) will do all the decompression on the video card, so all other things being equal the more dedicated GPU memory you have, the better off you are!

[0x80]

I wouldn't be so quick to assume that a discrete graphics card is always going to outperform an integrated solution like Iris Pro. If you've read the other link I pointed to, you can see that in some scenarios the integrated graphics performed as good as or even outperformed the discrete card.

This benchmark is mainly about how many videos can be decoded and rendered. If both GPUs decode 9 videos at 30fps, without a monitor attached, why would you expect it to be so much different if an external output is involved? And why would discrete graphics have it so much easier to drive this display? Just because it has more and faster memory? I'm curious to hear from the developers at Resolume if the DXV decoding would actually benefit from 2 vs 1 GB in the benchmark test. Because if the decoding happens frame by frame, and the compressed video files are not buffered in the GPU's RAM, then I don't expect there's a lot of memory required for 9 layers.

My theory so far is that the recent generations of integrated graphics like the Iris Pro is in a lot of ways catching up with discrete GPUs. And although the discrete solution might always outperform the other in demanding 3D rendering scenario's, you can't assume that every application is taxing the GPU in such ways that discrete graphics always have the upper hand.

Maybe I can't have hard evidence without having tested a computer with iris only graphics attached to an external screen, but the fact that my benchmarks clearly show no difference whatsoever in framerate, regardless of the external display, is at least suspicious to me.

[Oaktown]

I looked at the article you referred to in your first post and unless I'm missing something, the Unigine Heaven test corroborates my opinion, right?

Keep in mind that like most everyone in this forum, you'll want to connect more than 1 external display so the best way to know will for you to go test both systems with two or three 1920x1080 displays connected to the laptops.

[0x80]

Those benchmarks are all about 3D rendering performance. I'm in no way doubting that a discrete GPU is able to outperform an integrated solution. The only reason I supplied the article link was to show that there are also unexpected outcomes, because you can see there are a few tests where Iris Pro was able to keep up with the discrete GPU. In a Windows VM there was even a game which performed better on the Iris graphics. This has everything to do with how software is implemented and how the different features of drivers and GPU resources are used.

And this is why I think you shouldn't blindly trust that the 750M will outperform the Iris Pro by a long stretch in every scenario. Like when your GPU is mainly concerned with video decoding.

I still think I'm making a valid point here, and I am curious to know how Resolume and DXV in particular are using the GPU's resources.

[Oaktown]

I don't blindly trust the GT 750M, I use it all the time for shows and it performs amazingly well. All I'm saying is these tests you're referring are rather irrelevant since they are testing the Iris Pro vs the GT 750M using the Retina display of the Macbook Pro not an external display. I'll be curious to see the results when you get a chance to test two Macbook Pros (one with Iris Pro only and one with Iris Pro & GT 750M).

Maybe the Resolume team will chime in on the DXV codec and the use of GPU resources but you'll find plenty of threads on that topic if you use the forum search function.

[0x80]

I don't blindly trust the GT 750M, I use it all the time for shows and it performs amazingly well.

I didn't say that. All I'm saying is don't blindly trust it to always outperform the Iris Pro under all circumstances.

Maybe the Resolume team will chime in on the DXV codec and the use of GPU resources but you'll find plenty of threads on that topic if you use the forum search function.

I've tried searching for "dxv decoding" but found nothing about how GPU resources are used. Just that DXV decodes on the GPU. The only new thing I've learned is that ProRes might be a good alternative for Macs since it also decodes on the GPU. But if you can point me to more elaborate threads on this forums please do.

[Joris]

how GPU resources are used. Just that DXV decodes on the GPU

So what more would you like to know?

We use GLSL shaders to decompress the pixels in parallel instead of pixel by pixel. This is what makes DXV ( and HAP, and certain flavours of ProRes ) so much quicker to decode than other codecs.

We use standard GLSL / OpenGL calls for this. It's very likely that a modern Iris GPU can handle this type of operation just as well as a dedicated graphic card. The advise against using integrated cards is not so much because of performance issues, it's that they often do not support all the OpenGL extensions that Resolume uses. For instance, this can cause rendering artefacts on the older Intel HD series when warping the output in the Advanced Output.

It's our experience that the graphic cards in the latest Mac computers are categorically underperforming compared to running the same hardware under Windows. Like I said, we use standard OpenGL, which should be supported on every platform. If Apple all of a sudden requires OS specific optimisations, there's no documentation about this that we've heard of.

[0x80]

Thanks for chiming in Joris.

So what more would you like to know?

Oaktown suggested earlier that the 2GB dedicated memory vs 1GB shared memory of the Iris is also relevant. I replied that I don't expect it to make any difference unless DVX movies are buffered in large chunks on the GPU. That's why I was wondering about the implementation of the codec. If you are decoding frame by frame I don't see how more or even faster memory would make any difference.

We would have to do real world tests to be certain, but for now I hold on to my initial theory that the integrated graphics might have caught up to the discrete versions for certain applications, and that Resolume could well be one of them. Once I get hold of a late 2013 MBP without the 750M option I'll run the benchmark with an external display or two.

I have some hopes that OSX Yosemite will have improved OpenGL drivers

[hive8]

To put also some feedback in here (hope its not out of place), even if the openGL driver are optimized then they are for newer openGL version like 4+. some plugins made are still using older openGL code 2.1 and not the newer vertex and fragment shaders that we see on GSGL, so only RAW power will give you more power. More shading processors, more texture memory and so on. If a plugin or software doesn't take advantage of these newer calls then only RAW power will give you more performance. So i would choose a GT750 anytime over a IRIS Pro. and since we already on the GT860 versions that be even better, I really hope MAC will choose a GT860 or above for the upcoming model upgrade which should happen in the fall.